Digital Signal Processing, Год журнала: 2025, Номер unknown, С. 105307 - 105307
Опубликована: Май 1, 2025
Язык: Английский
Digital Signal Processing, Год журнала: 2025, Номер unknown, С. 105307 - 105307
Опубликована: Май 1, 2025
Язык: Английский
IEEE Transactions on Medical Imaging, Год журнала: 2023, Номер 43(2), С. 674 - 685
Опубликована: Сен. 19, 2023
Medical image segmentation and classification are two of the most key steps in computer-aided clinical diagnosis. The region interest were usually segmented a proper manner to extract useful features for further disease classification. However, these methods computationally complex time-consuming. In this paper, we proposed one-stage multi-task attention network (MTANet) which efficiently classifies objects an while generating high-quality mask each medical object. A reverse addition module was designed task fusion areas global map boundary cues high-resolution features, bottleneck used feature fusion. We evaluated performance MTANet with CNN-based transformer-based architectures across three imaging modalities different tasks: CVC-ClinicDB dataset polyp segmentation, ISIC-2018 skin lesion our private ultrasound liver tumor Our model outperformed state-of-the-art models on all datasets superior 25 radiologists
Язык: Английский
Процитировано
45IEEE Access, Год журнала: 2023, Номер 11, С. 16621 - 16630
Опубликована: Янв. 1, 2023
Medical image segmentation is a crucial way to assist doctors in the accurate diagnosis of diseases. However, accuracy medical needs further improvement due problems many noisy images and high similarity between background target regions. The current mainstream networks, such as TransUnet, have achieved segmentation. Still, encoders networks do not consider local connection adjacent chunks lack interaction inter-channel information during upsampling decoder. To address above problems, this paper proposed dual-encoder network, including HarDNet68 Transformer branch, which can extract features global feature input image, allowing network learn more information, thus improving effectiveness In paper, realize fusion different dimensions two stages encoding decoding, we propose adaptation module fuse channel multi-level channels, then improve accuracy. experimental results on CVC-ClinicDB, ETIS-Larib, COVID-19 CT datasets show that model performs better four evaluation metrics, Dice, Iou, Prec, Sens, achieves both internal filling edge prediction images. Accurate making critical cancerous regions advance, ensure cancer patients receive timely targeted treatment, their survival quality.
Язык: Английский
Процитировано
22IEEE Access, Год журнала: 2023, Номер 11, С. 95889 - 95902
Опубликована: Янв. 1, 2023
Colorectal cancer (CRC) is the third most common cause of cancer-related deaths in United States and anticipated to another 52,580 2023. The standard medical procedure for screening treating colorectal disease a colonoscopy. By effectively examining colonoscopy identify precancerous polyps early remove them before they become cancerous, CRC mortality can be lowered significantly. Manual examination detection time-consuming, tedious, prone human error. Automatic segmentation could fast practical; however, existing automated methods fail attain adequate accuracy segmentation. Moreover, these do not assess risk detected polyps. In this paper, we proposed an autonomous method detect their potential threats. utilized DoubleU-Net Vision Transformer (ViT) classifying based on risks. has achieved mean dice-coefficient 0.834 0.956 Endotech challenge Kvasir-SEG dataset, accordingly, outperforming state-of-the-art Then, classified segmented as hyper-plastic or adenomatous with 99% test accuracy.
Язык: Английский
Процитировано
19Bioengineering, Год журнала: 2024, Номер 11(3), С. 240 - 240
Опубликована: Фев. 28, 2024
This paper presents a novel U-Net model incorporating hybrid attention mechanism for automating the segmentation of sub-retinal layers in Optical Coherence Tomography (OCT) images. OCT is an ophthalmology tool that provides detailed insights into retinal structures. Manual these time-consuming and subjective, calling automated solutions. Our proposed combines edge spatial mechanisms with architecture to improve accuracy. By leveraging mechanisms, focuses selectively on image features. Extensive evaluations using datasets demonstrate our outperforms existing approaches, making it valuable medical professionals. The study also highlights model's robustness through performance metrics such as average Dice score 94.99%, Adjusted Rand Index (ARI) 97.00%, Strength Agreement (SOA) classifications like "Almost Perfect", "Excellent", "Very Strong". advanced predictive shows promise expediting processes enhancing precision ocular imaging real-world applications.
Язык: Английский
Процитировано
9Sensors, Год журнала: 2025, Номер 25(5), С. 1595 - 1595
Опубликована: Март 5, 2025
Due to low contrast, chromatic aberration, and generally small objects in underwater environments, a new fish detection model, BGLE-YOLO, is proposed investigate automated methods dedicated accurately detecting images. The model has parameters computational effort suitable for edge devices. First, an efficient multi-scale convolutional EMC module introduced enhance the backbone network capture dynamic changes targets environment. Secondly, global local feature fusion (BIG) integrated into neck preserve more information, reduce error information higher-level features, increase model’s effectiveness targets. Finally, prevent accuracy impact due excessive lightweighting, lightweight shared head (LSH) constructed. reparameterization technique further improves without additional cost. Experimental results of BGLE-YOLO on datasets DUO (Detection Underwater Objects) RUOD (Real-World Object Detection) show that achieves same as benchmark with ultra-low cost 6.2 GFLOPs parameter 1.6 MB.
Язык: Английский
Процитировано
1Applied Sciences, Год журнала: 2025, Номер 15(5), С. 2789 - 2789
Опубликована: Март 5, 2025
Unmanned aerial vehicle (UAV) imagery often suffers from significant object scale variations, high target density, and varying distances due to shooting conditions environmental factors, leading reduced robustness low detection accuracy in conventional models. To address these issues, this study adopts DGBL-YOLOv8s, an improved model tailored for UAV perspectives based on YOLOv8s. First, a Dilated Wide Residual (DWR) module is introduced replace the C2f backbone network of YOLOv8, enhancing model’s capability capture fine-grained features contextual information. Second, neck structure redesigned by incorporating Global-to-Local Spatial Aggregation (GLSA) combined with Bidirectional Feature Pyramid Network (BiFPN), which strengthens feature fusion. Third, lightweight shared convolution head proposed, batch normalization techniques. Additionally, further improve small detection, dedicated small-object introduced. Results experiments VisDrone dataset reveal that DGBL-YOLOv8s enhances 8.5% relative baseline model, alongside 34.8% reduction parameter count. The overall performance exceeds most current models, confirms advantages proposed improvement.
Язык: Английский
Процитировано
1Computers in Biology and Medicine, Год журнала: 2023, Номер 167, С. 107600 - 107600
Опубликована: Окт. 20, 2023
Язык: Английский
Процитировано
14Computers in Biology and Medicine, Год журнала: 2024, Номер 172, С. 108281 - 108281
Опубликована: Март 13, 2024
Язык: Английский
Процитировано
5Computers in Biology and Medicine, Год журнала: 2024, Номер 180, С. 108939 - 108939
Опубликована: Июль 29, 2024
Язык: Английский
Процитировано
5Cancers, Год журнала: 2024, Номер 16(6), С. 1120 - 1120
Опубликована: Март 11, 2024
Skin lesion segmentation plays a key role in the diagnosis of skin cancer; it can be component both traditional algorithms and end-to-end approaches. The quality directly impacts accuracy classification; however, attaining optimal necessitates substantial amount labeled data. Semi-supervised learning allows for employing unlabeled data to enhance results machine model. In case medical image segmentation, acquiring detailed annotation is time-consuming costly requires skilled individuals so utilization significant mitigation manual efforts. This study proposes novel approach semi-supervised using self-training with Noisy Student. utilizing large amounts available images. It consists four steps—first, training teacher model on only, then generating pseudo-labels model, student pseudo-labeled data, lastly, student* generated this work, we implemented DeepLabV3 architecture as models. As final result, achieved mIoU 88.0% ISIC 2018 dataset 87.54% PH2 dataset. evaluation proposed shows that Student improves performance neural networks task while only small
Язык: Английский
Процитировано
4